Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Extensible Firmware Interface
4 *
5 * Based on Extensible Firmware Interface Specification version 0.9
6 * April 30, 1999
7 *
8 * Copyright (C) 1999 VA Linux Systems
9 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
10 * Copyright (C) 1999-2003 Hewlett-Packard Co.
11 * David Mosberger-Tang <davidm@hpl.hp.com>
12 * Stephane Eranian <eranian@hpl.hp.com>
13 * (c) Copyright 2006 Hewlett-Packard Development Company, L.P.
14 * Bjorn Helgaas <bjorn.helgaas@hp.com>
15 *
16 * All EFI Runtime Services are not implemented yet as EFI only
17 * supports physical mode addressing on SoftSDV. This is to be fixed
18 * in a future version. --drummond 1999-07-20
19 *
20 * Implemented EFI runtime services and virtual mode calls. --davidm
21 *
22 * Goutham Rao: <goutham.rao@intel.com>
23 * Skip non-WB memory and ignore empty memory ranges.
24 */
25#include <linux/module.h>
26#include <linux/memblock.h>
27#include <linux/crash_dump.h>
28#include <linux/kernel.h>
29#include <linux/init.h>
30#include <linux/types.h>
31#include <linux/slab.h>
32#include <linux/time.h>
33#include <linux/efi.h>
34#include <linux/kexec.h>
35#include <linux/mm.h>
36
37#include <asm/io.h>
38#include <asm/kregs.h>
39#include <asm/meminit.h>
40#include <asm/processor.h>
41#include <asm/mca.h>
42#include <asm/setup.h>
43#include <asm/tlbflush.h>
44
45#define EFI_DEBUG 0
46
47#define ESI_TABLE_GUID \
48 EFI_GUID(0x43EA58DC, 0xCF28, 0x4b06, 0xB3, \
49 0x91, 0xB7, 0x50, 0x59, 0x34, 0x2B, 0xD4)
50
51static unsigned long mps_phys = EFI_INVALID_TABLE_ADDR;
52static __initdata unsigned long palo_phys;
53
54unsigned long __initdata esi_phys = EFI_INVALID_TABLE_ADDR;
55unsigned long hcdp_phys = EFI_INVALID_TABLE_ADDR;
56unsigned long sal_systab_phys = EFI_INVALID_TABLE_ADDR;
57
58static const efi_config_table_type_t arch_tables[] __initconst = {
59 {ESI_TABLE_GUID, &esi_phys, "ESI" },
60 {HCDP_TABLE_GUID, &hcdp_phys, "HCDP" },
61 {MPS_TABLE_GUID, &mps_phys, "MPS" },
62 {PROCESSOR_ABSTRACTION_LAYER_OVERWRITE_GUID, &palo_phys, "PALO" },
63 {SAL_SYSTEM_TABLE_GUID, &sal_systab_phys, "SALsystab" },
64 {},
65};
66
67extern efi_status_t efi_call_phys (void *, ...);
68
69static efi_runtime_services_t *runtime;
70static u64 mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL;
71
72#define efi_call_virt(f, args...) (*(f))(args)
73
74#define STUB_GET_TIME(prefix, adjust_arg) \
75static efi_status_t \
76prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \
77{ \
78 struct ia64_fpreg fr[6]; \
79 efi_time_cap_t *atc = NULL; \
80 efi_status_t ret; \
81 \
82 if (tc) \
83 atc = adjust_arg(tc); \
84 ia64_save_scratch_fpregs(fr); \
85 ret = efi_call_##prefix((efi_get_time_t *) __va(runtime->get_time), \
86 adjust_arg(tm), atc); \
87 ia64_load_scratch_fpregs(fr); \
88 return ret; \
89}
90
91#define STUB_SET_TIME(prefix, adjust_arg) \
92static efi_status_t \
93prefix##_set_time (efi_time_t *tm) \
94{ \
95 struct ia64_fpreg fr[6]; \
96 efi_status_t ret; \
97 \
98 ia64_save_scratch_fpregs(fr); \
99 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), \
100 adjust_arg(tm)); \
101 ia64_load_scratch_fpregs(fr); \
102 return ret; \
103}
104
105#define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \
106static efi_status_t \
107prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, \
108 efi_time_t *tm) \
109{ \
110 struct ia64_fpreg fr[6]; \
111 efi_status_t ret; \
112 \
113 ia64_save_scratch_fpregs(fr); \
114 ret = efi_call_##prefix( \
115 (efi_get_wakeup_time_t *) __va(runtime->get_wakeup_time), \
116 adjust_arg(enabled), adjust_arg(pending), adjust_arg(tm)); \
117 ia64_load_scratch_fpregs(fr); \
118 return ret; \
119}
120
121#define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \
122static efi_status_t \
123prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \
124{ \
125 struct ia64_fpreg fr[6]; \
126 efi_time_t *atm = NULL; \
127 efi_status_t ret; \
128 \
129 if (tm) \
130 atm = adjust_arg(tm); \
131 ia64_save_scratch_fpregs(fr); \
132 ret = efi_call_##prefix( \
133 (efi_set_wakeup_time_t *) __va(runtime->set_wakeup_time), \
134 enabled, atm); \
135 ia64_load_scratch_fpregs(fr); \
136 return ret; \
137}
138
139#define STUB_GET_VARIABLE(prefix, adjust_arg) \
140static efi_status_t \
141prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \
142 unsigned long *data_size, void *data) \
143{ \
144 struct ia64_fpreg fr[6]; \
145 u32 *aattr = NULL; \
146 efi_status_t ret; \
147 \
148 if (attr) \
149 aattr = adjust_arg(attr); \
150 ia64_save_scratch_fpregs(fr); \
151 ret = efi_call_##prefix( \
152 (efi_get_variable_t *) __va(runtime->get_variable), \
153 adjust_arg(name), adjust_arg(vendor), aattr, \
154 adjust_arg(data_size), adjust_arg(data)); \
155 ia64_load_scratch_fpregs(fr); \
156 return ret; \
157}
158
159#define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \
160static efi_status_t \
161prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, \
162 efi_guid_t *vendor) \
163{ \
164 struct ia64_fpreg fr[6]; \
165 efi_status_t ret; \
166 \
167 ia64_save_scratch_fpregs(fr); \
168 ret = efi_call_##prefix( \
169 (efi_get_next_variable_t *) __va(runtime->get_next_variable), \
170 adjust_arg(name_size), adjust_arg(name), adjust_arg(vendor)); \
171 ia64_load_scratch_fpregs(fr); \
172 return ret; \
173}
174
175#define STUB_SET_VARIABLE(prefix, adjust_arg) \
176static efi_status_t \
177prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, \
178 u32 attr, unsigned long data_size, \
179 void *data) \
180{ \
181 struct ia64_fpreg fr[6]; \
182 efi_status_t ret; \
183 \
184 ia64_save_scratch_fpregs(fr); \
185 ret = efi_call_##prefix( \
186 (efi_set_variable_t *) __va(runtime->set_variable), \
187 adjust_arg(name), adjust_arg(vendor), attr, data_size, \
188 adjust_arg(data)); \
189 ia64_load_scratch_fpregs(fr); \
190 return ret; \
191}
192
193#define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \
194static efi_status_t \
195prefix##_get_next_high_mono_count (u32 *count) \
196{ \
197 struct ia64_fpreg fr[6]; \
198 efi_status_t ret; \
199 \
200 ia64_save_scratch_fpregs(fr); \
201 ret = efi_call_##prefix((efi_get_next_high_mono_count_t *) \
202 __va(runtime->get_next_high_mono_count), \
203 adjust_arg(count)); \
204 ia64_load_scratch_fpregs(fr); \
205 return ret; \
206}
207
208#define STUB_RESET_SYSTEM(prefix, adjust_arg) \
209static void \
210prefix##_reset_system (int reset_type, efi_status_t status, \
211 unsigned long data_size, efi_char16_t *data) \
212{ \
213 struct ia64_fpreg fr[6]; \
214 efi_char16_t *adata = NULL; \
215 \
216 if (data) \
217 adata = adjust_arg(data); \
218 \
219 ia64_save_scratch_fpregs(fr); \
220 efi_call_##prefix( \
221 (efi_reset_system_t *) __va(runtime->reset_system), \
222 reset_type, status, data_size, adata); \
223 /* should not return, but just in case... */ \
224 ia64_load_scratch_fpregs(fr); \
225}
226
227#define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg))
228
229STUB_GET_TIME(phys, phys_ptr)
230STUB_SET_TIME(phys, phys_ptr)
231STUB_GET_WAKEUP_TIME(phys, phys_ptr)
232STUB_SET_WAKEUP_TIME(phys, phys_ptr)
233STUB_GET_VARIABLE(phys, phys_ptr)
234STUB_GET_NEXT_VARIABLE(phys, phys_ptr)
235STUB_SET_VARIABLE(phys, phys_ptr)
236STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr)
237STUB_RESET_SYSTEM(phys, phys_ptr)
238
239#define id(arg) arg
240
241STUB_GET_TIME(virt, id)
242STUB_SET_TIME(virt, id)
243STUB_GET_WAKEUP_TIME(virt, id)
244STUB_SET_WAKEUP_TIME(virt, id)
245STUB_GET_VARIABLE(virt, id)
246STUB_GET_NEXT_VARIABLE(virt, id)
247STUB_SET_VARIABLE(virt, id)
248STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id)
249STUB_RESET_SYSTEM(virt, id)
250
251void
252efi_gettimeofday (struct timespec64 *ts)
253{
254 efi_time_t tm;
255
256 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) {
257 memset(ts, 0, sizeof(*ts));
258 return;
259 }
260
261 ts->tv_sec = mktime64(tm.year, tm.month, tm.day,
262 tm.hour, tm.minute, tm.second);
263 ts->tv_nsec = tm.nanosecond;
264}
265
266static int
267is_memory_available (efi_memory_desc_t *md)
268{
269 if (!(md->attribute & EFI_MEMORY_WB))
270 return 0;
271
272 switch (md->type) {
273 case EFI_LOADER_CODE:
274 case EFI_LOADER_DATA:
275 case EFI_BOOT_SERVICES_CODE:
276 case EFI_BOOT_SERVICES_DATA:
277 case EFI_CONVENTIONAL_MEMORY:
278 return 1;
279 }
280 return 0;
281}
282
283typedef struct kern_memdesc {
284 u64 attribute;
285 u64 start;
286 u64 num_pages;
287} kern_memdesc_t;
288
289static kern_memdesc_t *kern_memmap;
290
291#define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
292
293static inline u64
294kmd_end(kern_memdesc_t *kmd)
295{
296 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
297}
298
299static inline u64
300efi_md_end(efi_memory_desc_t *md)
301{
302 return (md->phys_addr + efi_md_size(md));
303}
304
305static inline int
306efi_wb(efi_memory_desc_t *md)
307{
308 return (md->attribute & EFI_MEMORY_WB);
309}
310
311static inline int
312efi_uc(efi_memory_desc_t *md)
313{
314 return (md->attribute & EFI_MEMORY_UC);
315}
316
317static void
318walk (efi_freemem_callback_t callback, void *arg, u64 attr)
319{
320 kern_memdesc_t *k;
321 u64 start, end, voff;
322
323 voff = (attr == EFI_MEMORY_WB) ? PAGE_OFFSET : __IA64_UNCACHED_OFFSET;
324 for (k = kern_memmap; k->start != ~0UL; k++) {
325 if (k->attribute != attr)
326 continue;
327 start = PAGE_ALIGN(k->start);
328 end = (k->start + (k->num_pages << EFI_PAGE_SHIFT)) & PAGE_MASK;
329 if (start < end)
330 if ((*callback)(start + voff, end + voff, arg) < 0)
331 return;
332 }
333}
334
335/*
336 * Walk the EFI memory map and call CALLBACK once for each EFI memory
337 * descriptor that has memory that is available for OS use.
338 */
339void
340efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
341{
342 walk(callback, arg, EFI_MEMORY_WB);
343}
344
345/*
346 * Walk the EFI memory map and call CALLBACK once for each EFI memory
347 * descriptor that has memory that is available for uncached allocator.
348 */
349void
350efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
351{
352 walk(callback, arg, EFI_MEMORY_UC);
353}
354
355/*
356 * Look for the PAL_CODE region reported by EFI and map it using an
357 * ITR to enable safe PAL calls in virtual mode. See IA-64 Processor
358 * Abstraction Layer chapter 11 in ADAG
359 */
360void *
361efi_get_pal_addr (void)
362{
363 void *efi_map_start, *efi_map_end, *p;
364 efi_memory_desc_t *md;
365 u64 efi_desc_size;
366 int pal_code_count = 0;
367 u64 vaddr, mask;
368
369 efi_map_start = __va(ia64_boot_param->efi_memmap);
370 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
371 efi_desc_size = ia64_boot_param->efi_memdesc_size;
372
373 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
374 md = p;
375 if (md->type != EFI_PAL_CODE)
376 continue;
377
378 if (++pal_code_count > 1) {
379 printk(KERN_ERR "Too many EFI Pal Code memory ranges, "
380 "dropped @ %llx\n", md->phys_addr);
381 continue;
382 }
383 /*
384 * The only ITLB entry in region 7 that is used is the one
385 * installed by __start(). That entry covers a 64MB range.
386 */
387 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
388 vaddr = PAGE_OFFSET + md->phys_addr;
389
390 /*
391 * We must check that the PAL mapping won't overlap with the
392 * kernel mapping.
393 *
394 * PAL code is guaranteed to be aligned on a power of 2 between
395 * 4k and 256KB and that only one ITR is needed to map it. This
396 * implies that the PAL code is always aligned on its size,
397 * i.e., the closest matching page size supported by the TLB.
398 * Therefore PAL code is guaranteed never to cross a 64MB unless
399 * it is bigger than 64MB (very unlikely!). So for now the
400 * following test is enough to determine whether or not we need
401 * a dedicated ITR for the PAL code.
402 */
403 if ((vaddr & mask) == (KERNEL_START & mask)) {
404 printk(KERN_INFO "%s: no need to install ITR for PAL code\n",
405 __func__);
406 continue;
407 }
408
409 if (efi_md_size(md) > IA64_GRANULE_SIZE)
410 panic("Whoa! PAL code size bigger than a granule!");
411
412#if EFI_DEBUG
413 mask = ~((1 << IA64_GRANULE_SHIFT) - 1);
414
415 printk(KERN_INFO "CPU %d: mapping PAL code "
416 "[0x%lx-0x%lx) into [0x%lx-0x%lx)\n",
417 smp_processor_id(), md->phys_addr,
418 md->phys_addr + efi_md_size(md),
419 vaddr & mask, (vaddr & mask) + IA64_GRANULE_SIZE);
420#endif
421 return __va(md->phys_addr);
422 }
423 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n",
424 __func__);
425 return NULL;
426}
427
428
429static u8 __init palo_checksum(u8 *buffer, u32 length)
430{
431 u8 sum = 0;
432 u8 *end = buffer + length;
433
434 while (buffer < end)
435 sum = (u8) (sum + *(buffer++));
436
437 return sum;
438}
439
440/*
441 * Parse and handle PALO table which is published at:
442 * http://www.dig64.org/home/DIG64_PALO_R1_0.pdf
443 */
444static void __init handle_palo(unsigned long phys_addr)
445{
446 struct palo_table *palo = __va(phys_addr);
447 u8 checksum;
448
449 if (strncmp(palo->signature, PALO_SIG, sizeof(PALO_SIG) - 1)) {
450 printk(KERN_INFO "PALO signature incorrect.\n");
451 return;
452 }
453
454 checksum = palo_checksum((u8 *)palo, palo->length);
455 if (checksum) {
456 printk(KERN_INFO "PALO checksum incorrect.\n");
457 return;
458 }
459
460 setup_ptcg_sem(palo->max_tlb_purges, NPTCG_FROM_PALO);
461}
462
463void
464efi_map_pal_code (void)
465{
466 void *pal_vaddr = efi_get_pal_addr ();
467 u64 psr;
468
469 if (!pal_vaddr)
470 return;
471
472 /*
473 * Cannot write to CRx with PSR.ic=1
474 */
475 psr = ia64_clear_ic();
476 ia64_itr(0x1, IA64_TR_PALCODE,
477 GRANULEROUNDDOWN((unsigned long) pal_vaddr),
478 pte_val(pfn_pte(__pa(pal_vaddr) >> PAGE_SHIFT, PAGE_KERNEL)),
479 IA64_GRANULE_SHIFT);
480 ia64_set_psr(psr); /* restore psr */
481}
482
483void __init
484efi_init (void)
485{
486 const efi_system_table_t *efi_systab;
487 void *efi_map_start, *efi_map_end;
488 u64 efi_desc_size;
489 char *cp;
490
491 set_bit(EFI_BOOT, &efi.flags);
492 set_bit(EFI_64BIT, &efi.flags);
493
494 /*
495 * It's too early to be able to use the standard kernel command line
496 * support...
497 */
498 for (cp = boot_command_line; *cp; ) {
499 if (memcmp(cp, "mem=", 4) == 0) {
500 mem_limit = memparse(cp + 4, &cp);
501 } else if (memcmp(cp, "max_addr=", 9) == 0) {
502 max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
503 } else if (memcmp(cp, "min_addr=", 9) == 0) {
504 min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
505 } else {
506 while (*cp != ' ' && *cp)
507 ++cp;
508 while (*cp == ' ')
509 ++cp;
510 }
511 }
512 if (min_addr != 0UL)
513 printk(KERN_INFO "Ignoring memory below %lluMB\n",
514 min_addr >> 20);
515 if (max_addr != ~0UL)
516 printk(KERN_INFO "Ignoring memory above %lluMB\n",
517 max_addr >> 20);
518
519 efi_systab = __va(ia64_boot_param->efi_systab);
520
521 /*
522 * Verify the EFI Table
523 */
524 if (efi_systab == NULL)
525 panic("Whoa! Can't find EFI system table.\n");
526 if (efi_systab_check_header(&efi_systab->hdr, 1))
527 panic("Whoa! EFI system table signature incorrect\n");
528
529 efi_systab_report_header(&efi_systab->hdr, efi_systab->fw_vendor);
530
531 palo_phys = EFI_INVALID_TABLE_ADDR;
532
533 if (efi_config_parse_tables(__va(efi_systab->tables),
534 efi_systab->nr_tables,
535 arch_tables) != 0)
536 return;
537
538 if (palo_phys != EFI_INVALID_TABLE_ADDR)
539 handle_palo(palo_phys);
540
541 runtime = __va(efi_systab->runtime);
542 efi.get_time = phys_get_time;
543 efi.set_time = phys_set_time;
544 efi.get_wakeup_time = phys_get_wakeup_time;
545 efi.set_wakeup_time = phys_set_wakeup_time;
546 efi.get_variable = phys_get_variable;
547 efi.get_next_variable = phys_get_next_variable;
548 efi.set_variable = phys_set_variable;
549 efi.get_next_high_mono_count = phys_get_next_high_mono_count;
550 efi.reset_system = phys_reset_system;
551
552 efi_map_start = __va(ia64_boot_param->efi_memmap);
553 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
554 efi_desc_size = ia64_boot_param->efi_memdesc_size;
555
556#if EFI_DEBUG
557 /* print EFI memory map: */
558 {
559 efi_memory_desc_t *md;
560 void *p;
561
562 for (i = 0, p = efi_map_start; p < efi_map_end;
563 ++i, p += efi_desc_size)
564 {
565 const char *unit;
566 unsigned long size;
567 char buf[64];
568
569 md = p;
570 size = md->num_pages << EFI_PAGE_SHIFT;
571
572 if ((size >> 40) > 0) {
573 size >>= 40;
574 unit = "TB";
575 } else if ((size >> 30) > 0) {
576 size >>= 30;
577 unit = "GB";
578 } else if ((size >> 20) > 0) {
579 size >>= 20;
580 unit = "MB";
581 } else {
582 size >>= 10;
583 unit = "KB";
584 }
585
586 printk("mem%02d: %s "
587 "range=[0x%016lx-0x%016lx) (%4lu%s)\n",
588 i, efi_md_typeattr_format(buf, sizeof(buf), md),
589 md->phys_addr,
590 md->phys_addr + efi_md_size(md), size, unit);
591 }
592 }
593#endif
594
595 efi_map_pal_code();
596 efi_enter_virtual_mode();
597}
598
599void
600efi_enter_virtual_mode (void)
601{
602 void *efi_map_start, *efi_map_end, *p;
603 efi_memory_desc_t *md;
604 efi_status_t status;
605 u64 efi_desc_size;
606
607 efi_map_start = __va(ia64_boot_param->efi_memmap);
608 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
609 efi_desc_size = ia64_boot_param->efi_memdesc_size;
610
611 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
612 md = p;
613 if (md->attribute & EFI_MEMORY_RUNTIME) {
614 /*
615 * Some descriptors have multiple bits set, so the
616 * order of the tests is relevant.
617 */
618 if (md->attribute & EFI_MEMORY_WB) {
619 md->virt_addr = (u64) __va(md->phys_addr);
620 } else if (md->attribute & EFI_MEMORY_UC) {
621 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
622 } else if (md->attribute & EFI_MEMORY_WC) {
623#if 0
624 md->virt_addr = ia64_remap(md->phys_addr,
625 (_PAGE_A |
626 _PAGE_P |
627 _PAGE_D |
628 _PAGE_MA_WC |
629 _PAGE_PL_0 |
630 _PAGE_AR_RW));
631#else
632 printk(KERN_INFO "EFI_MEMORY_WC mapping\n");
633 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
634#endif
635 } else if (md->attribute & EFI_MEMORY_WT) {
636#if 0
637 md->virt_addr = ia64_remap(md->phys_addr,
638 (_PAGE_A |
639 _PAGE_P |
640 _PAGE_D |
641 _PAGE_MA_WT |
642 _PAGE_PL_0 |
643 _PAGE_AR_RW));
644#else
645 printk(KERN_INFO "EFI_MEMORY_WT mapping\n");
646 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
647#endif
648 }
649 }
650 }
651
652 status = efi_call_phys(__va(runtime->set_virtual_address_map),
653 ia64_boot_param->efi_memmap_size,
654 efi_desc_size,
655 ia64_boot_param->efi_memdesc_version,
656 ia64_boot_param->efi_memmap);
657 if (status != EFI_SUCCESS) {
658 printk(KERN_WARNING "warning: unable to switch EFI into "
659 "virtual mode (status=%lu)\n", status);
660 return;
661 }
662
663 set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
664
665 /*
666 * Now that EFI is in virtual mode, we call the EFI functions more
667 * efficiently:
668 */
669 efi.get_time = virt_get_time;
670 efi.set_time = virt_set_time;
671 efi.get_wakeup_time = virt_get_wakeup_time;
672 efi.set_wakeup_time = virt_set_wakeup_time;
673 efi.get_variable = virt_get_variable;
674 efi.get_next_variable = virt_get_next_variable;
675 efi.set_variable = virt_set_variable;
676 efi.get_next_high_mono_count = virt_get_next_high_mono_count;
677 efi.reset_system = virt_reset_system;
678}
679
680/*
681 * Walk the EFI memory map looking for the I/O port range. There can only be
682 * one entry of this type, other I/O port ranges should be described via ACPI.
683 */
684u64
685efi_get_iobase (void)
686{
687 void *efi_map_start, *efi_map_end, *p;
688 efi_memory_desc_t *md;
689 u64 efi_desc_size;
690
691 efi_map_start = __va(ia64_boot_param->efi_memmap);
692 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
693 efi_desc_size = ia64_boot_param->efi_memdesc_size;
694
695 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
696 md = p;
697 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) {
698 if (md->attribute & EFI_MEMORY_UC)
699 return md->phys_addr;
700 }
701 }
702 return 0;
703}
704
705static struct kern_memdesc *
706kern_memory_descriptor (unsigned long phys_addr)
707{
708 struct kern_memdesc *md;
709
710 for (md = kern_memmap; md->start != ~0UL; md++) {
711 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
712 return md;
713 }
714 return NULL;
715}
716
717static efi_memory_desc_t *
718efi_memory_descriptor (unsigned long phys_addr)
719{
720 void *efi_map_start, *efi_map_end, *p;
721 efi_memory_desc_t *md;
722 u64 efi_desc_size;
723
724 efi_map_start = __va(ia64_boot_param->efi_memmap);
725 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
726 efi_desc_size = ia64_boot_param->efi_memdesc_size;
727
728 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
729 md = p;
730
731 if (phys_addr - md->phys_addr < efi_md_size(md))
732 return md;
733 }
734 return NULL;
735}
736
737static int
738efi_memmap_intersects (unsigned long phys_addr, unsigned long size)
739{
740 void *efi_map_start, *efi_map_end, *p;
741 efi_memory_desc_t *md;
742 u64 efi_desc_size;
743 unsigned long end;
744
745 efi_map_start = __va(ia64_boot_param->efi_memmap);
746 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
747 efi_desc_size = ia64_boot_param->efi_memdesc_size;
748
749 end = phys_addr + size;
750
751 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
752 md = p;
753 if (md->phys_addr < end && efi_md_end(md) > phys_addr)
754 return 1;
755 }
756 return 0;
757}
758
759int
760efi_mem_type (unsigned long phys_addr)
761{
762 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
763
764 if (md)
765 return md->type;
766 return -EINVAL;
767}
768
769u64
770efi_mem_attributes (unsigned long phys_addr)
771{
772 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
773
774 if (md)
775 return md->attribute;
776 return 0;
777}
778EXPORT_SYMBOL(efi_mem_attributes);
779
780u64
781efi_mem_attribute (unsigned long phys_addr, unsigned long size)
782{
783 unsigned long end = phys_addr + size;
784 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
785 u64 attr;
786
787 if (!md)
788 return 0;
789
790 /*
791 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
792 * the kernel that firmware needs this region mapped.
793 */
794 attr = md->attribute & ~EFI_MEMORY_RUNTIME;
795 do {
796 unsigned long md_end = efi_md_end(md);
797
798 if (end <= md_end)
799 return attr;
800
801 md = efi_memory_descriptor(md_end);
802 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr)
803 return 0;
804 } while (md);
805 return 0; /* never reached */
806}
807
808u64
809kern_mem_attribute (unsigned long phys_addr, unsigned long size)
810{
811 unsigned long end = phys_addr + size;
812 struct kern_memdesc *md;
813 u64 attr;
814
815 /*
816 * This is a hack for ioremap calls before we set up kern_memmap.
817 * Maybe we should do efi_memmap_init() earlier instead.
818 */
819 if (!kern_memmap) {
820 attr = efi_mem_attribute(phys_addr, size);
821 if (attr & EFI_MEMORY_WB)
822 return EFI_MEMORY_WB;
823 return 0;
824 }
825
826 md = kern_memory_descriptor(phys_addr);
827 if (!md)
828 return 0;
829
830 attr = md->attribute;
831 do {
832 unsigned long md_end = kmd_end(md);
833
834 if (end <= md_end)
835 return attr;
836
837 md = kern_memory_descriptor(md_end);
838 if (!md || md->attribute != attr)
839 return 0;
840 } while (md);
841 return 0; /* never reached */
842}
843
844int
845valid_phys_addr_range (phys_addr_t phys_addr, unsigned long size)
846{
847 u64 attr;
848
849 /*
850 * /dev/mem reads and writes use copy_to_user(), which implicitly
851 * uses a granule-sized kernel identity mapping. It's really
852 * only safe to do this for regions in kern_memmap. For more
853 * details, see Documentation/ia64/aliasing.rst.
854 */
855 attr = kern_mem_attribute(phys_addr, size);
856 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
857 return 1;
858 return 0;
859}
860
861int
862valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size)
863{
864 unsigned long phys_addr = pfn << PAGE_SHIFT;
865 u64 attr;
866
867 attr = efi_mem_attribute(phys_addr, size);
868
869 /*
870 * /dev/mem mmap uses normal user pages, so we don't need the entire
871 * granule, but the entire region we're mapping must support the same
872 * attribute.
873 */
874 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
875 return 1;
876
877 /*
878 * Intel firmware doesn't tell us about all the MMIO regions, so
879 * in general we have to allow mmap requests. But if EFI *does*
880 * tell us about anything inside this region, we should deny it.
881 * The user can always map a smaller region to avoid the overlap.
882 */
883 if (efi_memmap_intersects(phys_addr, size))
884 return 0;
885
886 return 1;
887}
888
889pgprot_t
890phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size,
891 pgprot_t vma_prot)
892{
893 unsigned long phys_addr = pfn << PAGE_SHIFT;
894 u64 attr;
895
896 /*
897 * For /dev/mem mmap, we use user mappings, but if the region is
898 * in kern_memmap (and hence may be covered by a kernel mapping),
899 * we must use the same attribute as the kernel mapping.
900 */
901 attr = kern_mem_attribute(phys_addr, size);
902 if (attr & EFI_MEMORY_WB)
903 return pgprot_cacheable(vma_prot);
904 else if (attr & EFI_MEMORY_UC)
905 return pgprot_noncached(vma_prot);
906
907 /*
908 * Some chipsets don't support UC access to memory. If
909 * WB is supported, we prefer that.
910 */
911 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB)
912 return pgprot_cacheable(vma_prot);
913
914 return pgprot_noncached(vma_prot);
915}
916
917int __init
918efi_uart_console_only(void)
919{
920 efi_status_t status;
921 char *s, name[] = "ConOut";
922 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID;
923 efi_char16_t *utf16, name_utf16[32];
924 unsigned char data[1024];
925 unsigned long size = sizeof(data);
926 struct efi_generic_dev_path *hdr, *end_addr;
927 int uart = 0;
928
929 /* Convert to UTF-16 */
930 utf16 = name_utf16;
931 s = name;
932 while (*s)
933 *utf16++ = *s++ & 0x7f;
934 *utf16 = 0;
935
936 status = efi.get_variable(name_utf16, &guid, NULL, &size, data);
937 if (status != EFI_SUCCESS) {
938 printk(KERN_ERR "No EFI %s variable?\n", name);
939 return 0;
940 }
941
942 hdr = (struct efi_generic_dev_path *) data;
943 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size);
944 while (hdr < end_addr) {
945 if (hdr->type == EFI_DEV_MSG &&
946 hdr->sub_type == EFI_DEV_MSG_UART)
947 uart = 1;
948 else if (hdr->type == EFI_DEV_END_PATH ||
949 hdr->type == EFI_DEV_END_PATH2) {
950 if (!uart)
951 return 0;
952 if (hdr->sub_type == EFI_DEV_END_ENTIRE)
953 return 1;
954 uart = 0;
955 }
956 hdr = (struct efi_generic_dev_path *)((u8 *) hdr + hdr->length);
957 }
958 printk(KERN_ERR "Malformed %s value\n", name);
959 return 0;
960}
961
962/*
963 * Look for the first granule aligned memory descriptor memory
964 * that is big enough to hold EFI memory map. Make sure this
965 * descriptor is at least granule sized so it does not get trimmed
966 */
967struct kern_memdesc *
968find_memmap_space (void)
969{
970 u64 contig_low=0, contig_high=0;
971 u64 as = 0, ae;
972 void *efi_map_start, *efi_map_end, *p, *q;
973 efi_memory_desc_t *md, *pmd = NULL, *check_md;
974 u64 space_needed, efi_desc_size;
975 unsigned long total_mem = 0;
976
977 efi_map_start = __va(ia64_boot_param->efi_memmap);
978 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
979 efi_desc_size = ia64_boot_param->efi_memdesc_size;
980
981 /*
982 * Worst case: we need 3 kernel descriptors for each efi descriptor
983 * (if every entry has a WB part in the middle, and UC head and tail),
984 * plus one for the end marker.
985 */
986 space_needed = sizeof(kern_memdesc_t) *
987 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1);
988
989 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
990 md = p;
991 if (!efi_wb(md)) {
992 continue;
993 }
994 if (pmd == NULL || !efi_wb(pmd) ||
995 efi_md_end(pmd) != md->phys_addr) {
996 contig_low = GRANULEROUNDUP(md->phys_addr);
997 contig_high = efi_md_end(md);
998 for (q = p + efi_desc_size; q < efi_map_end;
999 q += efi_desc_size) {
1000 check_md = q;
1001 if (!efi_wb(check_md))
1002 break;
1003 if (contig_high != check_md->phys_addr)
1004 break;
1005 contig_high = efi_md_end(check_md);
1006 }
1007 contig_high = GRANULEROUNDDOWN(contig_high);
1008 }
1009 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA)
1010 continue;
1011
1012 /* Round ends inward to granule boundaries */
1013 as = max(contig_low, md->phys_addr);
1014 ae = min(contig_high, efi_md_end(md));
1015
1016 /* keep within max_addr= and min_addr= command line arg */
1017 as = max(as, min_addr);
1018 ae = min(ae, max_addr);
1019 if (ae <= as)
1020 continue;
1021
1022 /* avoid going over mem= command line arg */
1023 if (total_mem + (ae - as) > mem_limit)
1024 ae -= total_mem + (ae - as) - mem_limit;
1025
1026 if (ae <= as)
1027 continue;
1028
1029 if (ae - as > space_needed)
1030 break;
1031 }
1032 if (p >= efi_map_end)
1033 panic("Can't allocate space for kernel memory descriptors");
1034
1035 return __va(as);
1036}
1037
1038/*
1039 * Walk the EFI memory map and gather all memory available for kernel
1040 * to use. We can allocate partial granules only if the unavailable
1041 * parts exist, and are WB.
1042 */
1043unsigned long
1044efi_memmap_init(u64 *s, u64 *e)
1045{
1046 struct kern_memdesc *k, *prev = NULL;
1047 u64 contig_low=0, contig_high=0;
1048 u64 as, ae, lim;
1049 void *efi_map_start, *efi_map_end, *p, *q;
1050 efi_memory_desc_t *md, *pmd = NULL, *check_md;
1051 u64 efi_desc_size;
1052 unsigned long total_mem = 0;
1053
1054 k = kern_memmap = find_memmap_space();
1055
1056 efi_map_start = __va(ia64_boot_param->efi_memmap);
1057 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1058 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1059
1060 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
1061 md = p;
1062 if (!efi_wb(md)) {
1063 if (efi_uc(md) &&
1064 (md->type == EFI_CONVENTIONAL_MEMORY ||
1065 md->type == EFI_BOOT_SERVICES_DATA)) {
1066 k->attribute = EFI_MEMORY_UC;
1067 k->start = md->phys_addr;
1068 k->num_pages = md->num_pages;
1069 k++;
1070 }
1071 continue;
1072 }
1073 if (pmd == NULL || !efi_wb(pmd) ||
1074 efi_md_end(pmd) != md->phys_addr) {
1075 contig_low = GRANULEROUNDUP(md->phys_addr);
1076 contig_high = efi_md_end(md);
1077 for (q = p + efi_desc_size; q < efi_map_end;
1078 q += efi_desc_size) {
1079 check_md = q;
1080 if (!efi_wb(check_md))
1081 break;
1082 if (contig_high != check_md->phys_addr)
1083 break;
1084 contig_high = efi_md_end(check_md);
1085 }
1086 contig_high = GRANULEROUNDDOWN(contig_high);
1087 }
1088 if (!is_memory_available(md))
1089 continue;
1090
1091 /*
1092 * Round ends inward to granule boundaries
1093 * Give trimmings to uncached allocator
1094 */
1095 if (md->phys_addr < contig_low) {
1096 lim = min(efi_md_end(md), contig_low);
1097 if (efi_uc(md)) {
1098 if (k > kern_memmap &&
1099 (k-1)->attribute == EFI_MEMORY_UC &&
1100 kmd_end(k-1) == md->phys_addr) {
1101 (k-1)->num_pages +=
1102 (lim - md->phys_addr)
1103 >> EFI_PAGE_SHIFT;
1104 } else {
1105 k->attribute = EFI_MEMORY_UC;
1106 k->start = md->phys_addr;
1107 k->num_pages = (lim - md->phys_addr)
1108 >> EFI_PAGE_SHIFT;
1109 k++;
1110 }
1111 }
1112 as = contig_low;
1113 } else
1114 as = md->phys_addr;
1115
1116 if (efi_md_end(md) > contig_high) {
1117 lim = max(md->phys_addr, contig_high);
1118 if (efi_uc(md)) {
1119 if (lim == md->phys_addr && k > kern_memmap &&
1120 (k-1)->attribute == EFI_MEMORY_UC &&
1121 kmd_end(k-1) == md->phys_addr) {
1122 (k-1)->num_pages += md->num_pages;
1123 } else {
1124 k->attribute = EFI_MEMORY_UC;
1125 k->start = lim;
1126 k->num_pages = (efi_md_end(md) - lim)
1127 >> EFI_PAGE_SHIFT;
1128 k++;
1129 }
1130 }
1131 ae = contig_high;
1132 } else
1133 ae = efi_md_end(md);
1134
1135 /* keep within max_addr= and min_addr= command line arg */
1136 as = max(as, min_addr);
1137 ae = min(ae, max_addr);
1138 if (ae <= as)
1139 continue;
1140
1141 /* avoid going over mem= command line arg */
1142 if (total_mem + (ae - as) > mem_limit)
1143 ae -= total_mem + (ae - as) - mem_limit;
1144
1145 if (ae <= as)
1146 continue;
1147 if (prev && kmd_end(prev) == md->phys_addr) {
1148 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT;
1149 total_mem += ae - as;
1150 continue;
1151 }
1152 k->attribute = EFI_MEMORY_WB;
1153 k->start = as;
1154 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT;
1155 total_mem += ae - as;
1156 prev = k++;
1157 }
1158 k->start = ~0L; /* end-marker */
1159
1160 /* reserve the memory we are using for kern_memmap */
1161 *s = (u64)kern_memmap;
1162 *e = (u64)++k;
1163
1164 return total_mem;
1165}
1166
1167void
1168efi_initialize_iomem_resources(struct resource *code_resource,
1169 struct resource *data_resource,
1170 struct resource *bss_resource)
1171{
1172 struct resource *res;
1173 void *efi_map_start, *efi_map_end, *p;
1174 efi_memory_desc_t *md;
1175 u64 efi_desc_size;
1176 char *name;
1177 unsigned long flags, desc;
1178
1179 efi_map_start = __va(ia64_boot_param->efi_memmap);
1180 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1181 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1182
1183 res = NULL;
1184
1185 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1186 md = p;
1187
1188 if (md->num_pages == 0) /* should not happen */
1189 continue;
1190
1191 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1192 desc = IORES_DESC_NONE;
1193
1194 switch (md->type) {
1195
1196 case EFI_MEMORY_MAPPED_IO:
1197 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1198 continue;
1199
1200 case EFI_LOADER_CODE:
1201 case EFI_LOADER_DATA:
1202 case EFI_BOOT_SERVICES_DATA:
1203 case EFI_BOOT_SERVICES_CODE:
1204 case EFI_CONVENTIONAL_MEMORY:
1205 if (md->attribute & EFI_MEMORY_WP) {
1206 name = "System ROM";
1207 flags |= IORESOURCE_READONLY;
1208 } else if (md->attribute == EFI_MEMORY_UC) {
1209 name = "Uncached RAM";
1210 } else {
1211 name = "System RAM";
1212 flags |= IORESOURCE_SYSRAM;
1213 }
1214 break;
1215
1216 case EFI_ACPI_MEMORY_NVS:
1217 name = "ACPI Non-volatile Storage";
1218 desc = IORES_DESC_ACPI_NV_STORAGE;
1219 break;
1220
1221 case EFI_UNUSABLE_MEMORY:
1222 name = "reserved";
1223 flags |= IORESOURCE_DISABLED;
1224 break;
1225
1226 case EFI_PERSISTENT_MEMORY:
1227 name = "Persistent Memory";
1228 desc = IORES_DESC_PERSISTENT_MEMORY;
1229 break;
1230
1231 case EFI_RESERVED_TYPE:
1232 case EFI_RUNTIME_SERVICES_CODE:
1233 case EFI_RUNTIME_SERVICES_DATA:
1234 case EFI_ACPI_RECLAIM_MEMORY:
1235 default:
1236 name = "reserved";
1237 break;
1238 }
1239
1240 if ((res = kzalloc(sizeof(struct resource),
1241 GFP_KERNEL)) == NULL) {
1242 printk(KERN_ERR
1243 "failed to allocate resource for iomem\n");
1244 return;
1245 }
1246
1247 res->name = name;
1248 res->start = md->phys_addr;
1249 res->end = md->phys_addr + efi_md_size(md) - 1;
1250 res->flags = flags;
1251 res->desc = desc;
1252
1253 if (insert_resource(&iomem_resource, res) < 0)
1254 kfree(res);
1255 else {
1256 /*
1257 * We don't know which region contains
1258 * kernel data so we try it repeatedly and
1259 * let the resource manager test it.
1260 */
1261 insert_resource(res, code_resource);
1262 insert_resource(res, data_resource);
1263 insert_resource(res, bss_resource);
1264#ifdef CONFIG_KEXEC
1265 insert_resource(res, &efi_memmap_res);
1266 insert_resource(res, &boot_param_res);
1267 if (crashk_res.end > crashk_res.start)
1268 insert_resource(res, &crashk_res);
1269#endif
1270 }
1271 }
1272}
1273
1274#ifdef CONFIG_KEXEC
1275/* find a block of memory aligned to 64M exclude reserved regions
1276 rsvd_regions are sorted
1277 */
1278unsigned long __init
1279kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n)
1280{
1281 int i;
1282 u64 start, end;
1283 u64 alignment = 1UL << _PAGE_SIZE_64M;
1284 void *efi_map_start, *efi_map_end, *p;
1285 efi_memory_desc_t *md;
1286 u64 efi_desc_size;
1287
1288 efi_map_start = __va(ia64_boot_param->efi_memmap);
1289 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1290 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1291
1292 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1293 md = p;
1294 if (!efi_wb(md))
1295 continue;
1296 start = ALIGN(md->phys_addr, alignment);
1297 end = efi_md_end(md);
1298 for (i = 0; i < n; i++) {
1299 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) {
1300 if (__pa(r[i].start) > start + size)
1301 return start;
1302 start = ALIGN(__pa(r[i].end), alignment);
1303 if (i < n-1 &&
1304 __pa(r[i+1].start) < start + size)
1305 continue;
1306 else
1307 break;
1308 }
1309 }
1310 if (end > start + size)
1311 return start;
1312 }
1313
1314 printk(KERN_WARNING
1315 "Cannot reserve 0x%lx byte of memory for crashdump\n", size);
1316 return ~0UL;
1317}
1318#endif
1319
1320#ifdef CONFIG_CRASH_DUMP
1321/* locate the size find a the descriptor at a certain address */
1322unsigned long __init
1323vmcore_find_descriptor_size (unsigned long address)
1324{
1325 void *efi_map_start, *efi_map_end, *p;
1326 efi_memory_desc_t *md;
1327 u64 efi_desc_size;
1328 unsigned long ret = 0;
1329
1330 efi_map_start = __va(ia64_boot_param->efi_memmap);
1331 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1332 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1333
1334 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1335 md = p;
1336 if (efi_wb(md) && md->type == EFI_LOADER_DATA
1337 && md->phys_addr == address) {
1338 ret = efi_md_size(md);
1339 break;
1340 }
1341 }
1342
1343 if (ret == 0)
1344 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n");
1345
1346 return ret;
1347}
1348#endif
1349
1350char *efi_systab_show_arch(char *str)
1351{
1352 if (mps_phys != EFI_INVALID_TABLE_ADDR)
1353 str += sprintf(str, "MPS=0x%lx\n", mps_phys);
1354 if (hcdp_phys != EFI_INVALID_TABLE_ADDR)
1355 str += sprintf(str, "HCDP=0x%lx\n", hcdp_phys);
1356 return str;
1357}
1/*
2 * Extensible Firmware Interface
3 *
4 * Based on Extensible Firmware Interface Specification version 0.9
5 * April 30, 1999
6 *
7 * Copyright (C) 1999 VA Linux Systems
8 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
9 * Copyright (C) 1999-2003 Hewlett-Packard Co.
10 * David Mosberger-Tang <davidm@hpl.hp.com>
11 * Stephane Eranian <eranian@hpl.hp.com>
12 * (c) Copyright 2006 Hewlett-Packard Development Company, L.P.
13 * Bjorn Helgaas <bjorn.helgaas@hp.com>
14 *
15 * All EFI Runtime Services are not implemented yet as EFI only
16 * supports physical mode addressing on SoftSDV. This is to be fixed
17 * in a future version. --drummond 1999-07-20
18 *
19 * Implemented EFI runtime services and virtual mode calls. --davidm
20 *
21 * Goutham Rao: <goutham.rao@intel.com>
22 * Skip non-WB memory and ignore empty memory ranges.
23 */
24#include <linux/module.h>
25#include <linux/bootmem.h>
26#include <linux/crash_dump.h>
27#include <linux/kernel.h>
28#include <linux/init.h>
29#include <linux/types.h>
30#include <linux/slab.h>
31#include <linux/time.h>
32#include <linux/efi.h>
33#include <linux/kexec.h>
34#include <linux/mm.h>
35
36#include <asm/io.h>
37#include <asm/kregs.h>
38#include <asm/meminit.h>
39#include <asm/pgtable.h>
40#include <asm/processor.h>
41#include <asm/mca.h>
42#include <asm/setup.h>
43#include <asm/tlbflush.h>
44
45#define EFI_DEBUG 0
46
47static __initdata unsigned long palo_phys;
48
49static __initdata efi_config_table_type_t arch_tables[] = {
50 {PROCESSOR_ABSTRACTION_LAYER_OVERWRITE_GUID, "PALO", &palo_phys},
51 {NULL_GUID, NULL, 0},
52};
53
54extern efi_status_t efi_call_phys (void *, ...);
55
56static efi_runtime_services_t *runtime;
57static u64 mem_limit = ~0UL, max_addr = ~0UL, min_addr = 0UL;
58
59#define efi_call_virt(f, args...) (*(f))(args)
60
61#define STUB_GET_TIME(prefix, adjust_arg) \
62static efi_status_t \
63prefix##_get_time (efi_time_t *tm, efi_time_cap_t *tc) \
64{ \
65 struct ia64_fpreg fr[6]; \
66 efi_time_cap_t *atc = NULL; \
67 efi_status_t ret; \
68 \
69 if (tc) \
70 atc = adjust_arg(tc); \
71 ia64_save_scratch_fpregs(fr); \
72 ret = efi_call_##prefix((efi_get_time_t *) __va(runtime->get_time), \
73 adjust_arg(tm), atc); \
74 ia64_load_scratch_fpregs(fr); \
75 return ret; \
76}
77
78#define STUB_SET_TIME(prefix, adjust_arg) \
79static efi_status_t \
80prefix##_set_time (efi_time_t *tm) \
81{ \
82 struct ia64_fpreg fr[6]; \
83 efi_status_t ret; \
84 \
85 ia64_save_scratch_fpregs(fr); \
86 ret = efi_call_##prefix((efi_set_time_t *) __va(runtime->set_time), \
87 adjust_arg(tm)); \
88 ia64_load_scratch_fpregs(fr); \
89 return ret; \
90}
91
92#define STUB_GET_WAKEUP_TIME(prefix, adjust_arg) \
93static efi_status_t \
94prefix##_get_wakeup_time (efi_bool_t *enabled, efi_bool_t *pending, \
95 efi_time_t *tm) \
96{ \
97 struct ia64_fpreg fr[6]; \
98 efi_status_t ret; \
99 \
100 ia64_save_scratch_fpregs(fr); \
101 ret = efi_call_##prefix( \
102 (efi_get_wakeup_time_t *) __va(runtime->get_wakeup_time), \
103 adjust_arg(enabled), adjust_arg(pending), adjust_arg(tm)); \
104 ia64_load_scratch_fpregs(fr); \
105 return ret; \
106}
107
108#define STUB_SET_WAKEUP_TIME(prefix, adjust_arg) \
109static efi_status_t \
110prefix##_set_wakeup_time (efi_bool_t enabled, efi_time_t *tm) \
111{ \
112 struct ia64_fpreg fr[6]; \
113 efi_time_t *atm = NULL; \
114 efi_status_t ret; \
115 \
116 if (tm) \
117 atm = adjust_arg(tm); \
118 ia64_save_scratch_fpregs(fr); \
119 ret = efi_call_##prefix( \
120 (efi_set_wakeup_time_t *) __va(runtime->set_wakeup_time), \
121 enabled, atm); \
122 ia64_load_scratch_fpregs(fr); \
123 return ret; \
124}
125
126#define STUB_GET_VARIABLE(prefix, adjust_arg) \
127static efi_status_t \
128prefix##_get_variable (efi_char16_t *name, efi_guid_t *vendor, u32 *attr, \
129 unsigned long *data_size, void *data) \
130{ \
131 struct ia64_fpreg fr[6]; \
132 u32 *aattr = NULL; \
133 efi_status_t ret; \
134 \
135 if (attr) \
136 aattr = adjust_arg(attr); \
137 ia64_save_scratch_fpregs(fr); \
138 ret = efi_call_##prefix( \
139 (efi_get_variable_t *) __va(runtime->get_variable), \
140 adjust_arg(name), adjust_arg(vendor), aattr, \
141 adjust_arg(data_size), adjust_arg(data)); \
142 ia64_load_scratch_fpregs(fr); \
143 return ret; \
144}
145
146#define STUB_GET_NEXT_VARIABLE(prefix, adjust_arg) \
147static efi_status_t \
148prefix##_get_next_variable (unsigned long *name_size, efi_char16_t *name, \
149 efi_guid_t *vendor) \
150{ \
151 struct ia64_fpreg fr[6]; \
152 efi_status_t ret; \
153 \
154 ia64_save_scratch_fpregs(fr); \
155 ret = efi_call_##prefix( \
156 (efi_get_next_variable_t *) __va(runtime->get_next_variable), \
157 adjust_arg(name_size), adjust_arg(name), adjust_arg(vendor)); \
158 ia64_load_scratch_fpregs(fr); \
159 return ret; \
160}
161
162#define STUB_SET_VARIABLE(prefix, adjust_arg) \
163static efi_status_t \
164prefix##_set_variable (efi_char16_t *name, efi_guid_t *vendor, \
165 u32 attr, unsigned long data_size, \
166 void *data) \
167{ \
168 struct ia64_fpreg fr[6]; \
169 efi_status_t ret; \
170 \
171 ia64_save_scratch_fpregs(fr); \
172 ret = efi_call_##prefix( \
173 (efi_set_variable_t *) __va(runtime->set_variable), \
174 adjust_arg(name), adjust_arg(vendor), attr, data_size, \
175 adjust_arg(data)); \
176 ia64_load_scratch_fpregs(fr); \
177 return ret; \
178}
179
180#define STUB_GET_NEXT_HIGH_MONO_COUNT(prefix, adjust_arg) \
181static efi_status_t \
182prefix##_get_next_high_mono_count (u32 *count) \
183{ \
184 struct ia64_fpreg fr[6]; \
185 efi_status_t ret; \
186 \
187 ia64_save_scratch_fpregs(fr); \
188 ret = efi_call_##prefix((efi_get_next_high_mono_count_t *) \
189 __va(runtime->get_next_high_mono_count), \
190 adjust_arg(count)); \
191 ia64_load_scratch_fpregs(fr); \
192 return ret; \
193}
194
195#define STUB_RESET_SYSTEM(prefix, adjust_arg) \
196static void \
197prefix##_reset_system (int reset_type, efi_status_t status, \
198 unsigned long data_size, efi_char16_t *data) \
199{ \
200 struct ia64_fpreg fr[6]; \
201 efi_char16_t *adata = NULL; \
202 \
203 if (data) \
204 adata = adjust_arg(data); \
205 \
206 ia64_save_scratch_fpregs(fr); \
207 efi_call_##prefix( \
208 (efi_reset_system_t *) __va(runtime->reset_system), \
209 reset_type, status, data_size, adata); \
210 /* should not return, but just in case... */ \
211 ia64_load_scratch_fpregs(fr); \
212}
213
214#define phys_ptr(arg) ((__typeof__(arg)) ia64_tpa(arg))
215
216STUB_GET_TIME(phys, phys_ptr)
217STUB_SET_TIME(phys, phys_ptr)
218STUB_GET_WAKEUP_TIME(phys, phys_ptr)
219STUB_SET_WAKEUP_TIME(phys, phys_ptr)
220STUB_GET_VARIABLE(phys, phys_ptr)
221STUB_GET_NEXT_VARIABLE(phys, phys_ptr)
222STUB_SET_VARIABLE(phys, phys_ptr)
223STUB_GET_NEXT_HIGH_MONO_COUNT(phys, phys_ptr)
224STUB_RESET_SYSTEM(phys, phys_ptr)
225
226#define id(arg) arg
227
228STUB_GET_TIME(virt, id)
229STUB_SET_TIME(virt, id)
230STUB_GET_WAKEUP_TIME(virt, id)
231STUB_SET_WAKEUP_TIME(virt, id)
232STUB_GET_VARIABLE(virt, id)
233STUB_GET_NEXT_VARIABLE(virt, id)
234STUB_SET_VARIABLE(virt, id)
235STUB_GET_NEXT_HIGH_MONO_COUNT(virt, id)
236STUB_RESET_SYSTEM(virt, id)
237
238void
239efi_gettimeofday (struct timespec *ts)
240{
241 efi_time_t tm;
242
243 if ((*efi.get_time)(&tm, NULL) != EFI_SUCCESS) {
244 memset(ts, 0, sizeof(*ts));
245 return;
246 }
247
248 ts->tv_sec = mktime(tm.year, tm.month, tm.day,
249 tm.hour, tm.minute, tm.second);
250 ts->tv_nsec = tm.nanosecond;
251}
252
253static int
254is_memory_available (efi_memory_desc_t *md)
255{
256 if (!(md->attribute & EFI_MEMORY_WB))
257 return 0;
258
259 switch (md->type) {
260 case EFI_LOADER_CODE:
261 case EFI_LOADER_DATA:
262 case EFI_BOOT_SERVICES_CODE:
263 case EFI_BOOT_SERVICES_DATA:
264 case EFI_CONVENTIONAL_MEMORY:
265 return 1;
266 }
267 return 0;
268}
269
270typedef struct kern_memdesc {
271 u64 attribute;
272 u64 start;
273 u64 num_pages;
274} kern_memdesc_t;
275
276static kern_memdesc_t *kern_memmap;
277
278#define efi_md_size(md) (md->num_pages << EFI_PAGE_SHIFT)
279
280static inline u64
281kmd_end(kern_memdesc_t *kmd)
282{
283 return (kmd->start + (kmd->num_pages << EFI_PAGE_SHIFT));
284}
285
286static inline u64
287efi_md_end(efi_memory_desc_t *md)
288{
289 return (md->phys_addr + efi_md_size(md));
290}
291
292static inline int
293efi_wb(efi_memory_desc_t *md)
294{
295 return (md->attribute & EFI_MEMORY_WB);
296}
297
298static inline int
299efi_uc(efi_memory_desc_t *md)
300{
301 return (md->attribute & EFI_MEMORY_UC);
302}
303
304static void
305walk (efi_freemem_callback_t callback, void *arg, u64 attr)
306{
307 kern_memdesc_t *k;
308 u64 start, end, voff;
309
310 voff = (attr == EFI_MEMORY_WB) ? PAGE_OFFSET : __IA64_UNCACHED_OFFSET;
311 for (k = kern_memmap; k->start != ~0UL; k++) {
312 if (k->attribute != attr)
313 continue;
314 start = PAGE_ALIGN(k->start);
315 end = (k->start + (k->num_pages << EFI_PAGE_SHIFT)) & PAGE_MASK;
316 if (start < end)
317 if ((*callback)(start + voff, end + voff, arg) < 0)
318 return;
319 }
320}
321
322/*
323 * Walk the EFI memory map and call CALLBACK once for each EFI memory
324 * descriptor that has memory that is available for OS use.
325 */
326void
327efi_memmap_walk (efi_freemem_callback_t callback, void *arg)
328{
329 walk(callback, arg, EFI_MEMORY_WB);
330}
331
332/*
333 * Walk the EFI memory map and call CALLBACK once for each EFI memory
334 * descriptor that has memory that is available for uncached allocator.
335 */
336void
337efi_memmap_walk_uc (efi_freemem_callback_t callback, void *arg)
338{
339 walk(callback, arg, EFI_MEMORY_UC);
340}
341
342/*
343 * Look for the PAL_CODE region reported by EFI and map it using an
344 * ITR to enable safe PAL calls in virtual mode. See IA-64 Processor
345 * Abstraction Layer chapter 11 in ADAG
346 */
347void *
348efi_get_pal_addr (void)
349{
350 void *efi_map_start, *efi_map_end, *p;
351 efi_memory_desc_t *md;
352 u64 efi_desc_size;
353 int pal_code_count = 0;
354 u64 vaddr, mask;
355
356 efi_map_start = __va(ia64_boot_param->efi_memmap);
357 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
358 efi_desc_size = ia64_boot_param->efi_memdesc_size;
359
360 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
361 md = p;
362 if (md->type != EFI_PAL_CODE)
363 continue;
364
365 if (++pal_code_count > 1) {
366 printk(KERN_ERR "Too many EFI Pal Code memory ranges, "
367 "dropped @ %llx\n", md->phys_addr);
368 continue;
369 }
370 /*
371 * The only ITLB entry in region 7 that is used is the one
372 * installed by __start(). That entry covers a 64MB range.
373 */
374 mask = ~((1 << KERNEL_TR_PAGE_SHIFT) - 1);
375 vaddr = PAGE_OFFSET + md->phys_addr;
376
377 /*
378 * We must check that the PAL mapping won't overlap with the
379 * kernel mapping.
380 *
381 * PAL code is guaranteed to be aligned on a power of 2 between
382 * 4k and 256KB and that only one ITR is needed to map it. This
383 * implies that the PAL code is always aligned on its size,
384 * i.e., the closest matching page size supported by the TLB.
385 * Therefore PAL code is guaranteed never to cross a 64MB unless
386 * it is bigger than 64MB (very unlikely!). So for now the
387 * following test is enough to determine whether or not we need
388 * a dedicated ITR for the PAL code.
389 */
390 if ((vaddr & mask) == (KERNEL_START & mask)) {
391 printk(KERN_INFO "%s: no need to install ITR for PAL code\n",
392 __func__);
393 continue;
394 }
395
396 if (efi_md_size(md) > IA64_GRANULE_SIZE)
397 panic("Whoa! PAL code size bigger than a granule!");
398
399#if EFI_DEBUG
400 mask = ~((1 << IA64_GRANULE_SHIFT) - 1);
401
402 printk(KERN_INFO "CPU %d: mapping PAL code "
403 "[0x%lx-0x%lx) into [0x%lx-0x%lx)\n",
404 smp_processor_id(), md->phys_addr,
405 md->phys_addr + efi_md_size(md),
406 vaddr & mask, (vaddr & mask) + IA64_GRANULE_SIZE);
407#endif
408 return __va(md->phys_addr);
409 }
410 printk(KERN_WARNING "%s: no PAL-code memory-descriptor found\n",
411 __func__);
412 return NULL;
413}
414
415
416static u8 __init palo_checksum(u8 *buffer, u32 length)
417{
418 u8 sum = 0;
419 u8 *end = buffer + length;
420
421 while (buffer < end)
422 sum = (u8) (sum + *(buffer++));
423
424 return sum;
425}
426
427/*
428 * Parse and handle PALO table which is published at:
429 * http://www.dig64.org/home/DIG64_PALO_R1_0.pdf
430 */
431static void __init handle_palo(unsigned long phys_addr)
432{
433 struct palo_table *palo = __va(phys_addr);
434 u8 checksum;
435
436 if (strncmp(palo->signature, PALO_SIG, sizeof(PALO_SIG) - 1)) {
437 printk(KERN_INFO "PALO signature incorrect.\n");
438 return;
439 }
440
441 checksum = palo_checksum((u8 *)palo, palo->length);
442 if (checksum) {
443 printk(KERN_INFO "PALO checksum incorrect.\n");
444 return;
445 }
446
447 setup_ptcg_sem(palo->max_tlb_purges, NPTCG_FROM_PALO);
448}
449
450void
451efi_map_pal_code (void)
452{
453 void *pal_vaddr = efi_get_pal_addr ();
454 u64 psr;
455
456 if (!pal_vaddr)
457 return;
458
459 /*
460 * Cannot write to CRx with PSR.ic=1
461 */
462 psr = ia64_clear_ic();
463 ia64_itr(0x1, IA64_TR_PALCODE,
464 GRANULEROUNDDOWN((unsigned long) pal_vaddr),
465 pte_val(pfn_pte(__pa(pal_vaddr) >> PAGE_SHIFT, PAGE_KERNEL)),
466 IA64_GRANULE_SHIFT);
467 ia64_set_psr(psr); /* restore psr */
468}
469
470void __init
471efi_init (void)
472{
473 void *efi_map_start, *efi_map_end;
474 efi_char16_t *c16;
475 u64 efi_desc_size;
476 char *cp, vendor[100] = "unknown";
477 int i;
478
479 set_bit(EFI_BOOT, &efi.flags);
480 set_bit(EFI_64BIT, &efi.flags);
481
482 /*
483 * It's too early to be able to use the standard kernel command line
484 * support...
485 */
486 for (cp = boot_command_line; *cp; ) {
487 if (memcmp(cp, "mem=", 4) == 0) {
488 mem_limit = memparse(cp + 4, &cp);
489 } else if (memcmp(cp, "max_addr=", 9) == 0) {
490 max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
491 } else if (memcmp(cp, "min_addr=", 9) == 0) {
492 min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
493 } else {
494 while (*cp != ' ' && *cp)
495 ++cp;
496 while (*cp == ' ')
497 ++cp;
498 }
499 }
500 if (min_addr != 0UL)
501 printk(KERN_INFO "Ignoring memory below %lluMB\n",
502 min_addr >> 20);
503 if (max_addr != ~0UL)
504 printk(KERN_INFO "Ignoring memory above %lluMB\n",
505 max_addr >> 20);
506
507 efi.systab = __va(ia64_boot_param->efi_systab);
508
509 /*
510 * Verify the EFI Table
511 */
512 if (efi.systab == NULL)
513 panic("Whoa! Can't find EFI system table.\n");
514 if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE)
515 panic("Whoa! EFI system table signature incorrect\n");
516 if ((efi.systab->hdr.revision >> 16) == 0)
517 printk(KERN_WARNING "Warning: EFI system table version "
518 "%d.%02d, expected 1.00 or greater\n",
519 efi.systab->hdr.revision >> 16,
520 efi.systab->hdr.revision & 0xffff);
521
522 /* Show what we know for posterity */
523 c16 = __va(efi.systab->fw_vendor);
524 if (c16) {
525 for (i = 0;i < (int) sizeof(vendor) - 1 && *c16; ++i)
526 vendor[i] = *c16++;
527 vendor[i] = '\0';
528 }
529
530 printk(KERN_INFO "EFI v%u.%.02u by %s:",
531 efi.systab->hdr.revision >> 16,
532 efi.systab->hdr.revision & 0xffff, vendor);
533
534 set_bit(EFI_SYSTEM_TABLES, &efi.flags);
535
536 palo_phys = EFI_INVALID_TABLE_ADDR;
537
538 if (efi_config_init(arch_tables) != 0)
539 return;
540
541 if (palo_phys != EFI_INVALID_TABLE_ADDR)
542 handle_palo(palo_phys);
543
544 runtime = __va(efi.systab->runtime);
545 efi.get_time = phys_get_time;
546 efi.set_time = phys_set_time;
547 efi.get_wakeup_time = phys_get_wakeup_time;
548 efi.set_wakeup_time = phys_set_wakeup_time;
549 efi.get_variable = phys_get_variable;
550 efi.get_next_variable = phys_get_next_variable;
551 efi.set_variable = phys_set_variable;
552 efi.get_next_high_mono_count = phys_get_next_high_mono_count;
553 efi.reset_system = phys_reset_system;
554
555 efi_map_start = __va(ia64_boot_param->efi_memmap);
556 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
557 efi_desc_size = ia64_boot_param->efi_memdesc_size;
558
559#if EFI_DEBUG
560 /* print EFI memory map: */
561 {
562 efi_memory_desc_t *md;
563 void *p;
564
565 for (i = 0, p = efi_map_start; p < efi_map_end;
566 ++i, p += efi_desc_size)
567 {
568 const char *unit;
569 unsigned long size;
570 char buf[64];
571
572 md = p;
573 size = md->num_pages << EFI_PAGE_SHIFT;
574
575 if ((size >> 40) > 0) {
576 size >>= 40;
577 unit = "TB";
578 } else if ((size >> 30) > 0) {
579 size >>= 30;
580 unit = "GB";
581 } else if ((size >> 20) > 0) {
582 size >>= 20;
583 unit = "MB";
584 } else {
585 size >>= 10;
586 unit = "KB";
587 }
588
589 printk("mem%02d: %s "
590 "range=[0x%016lx-0x%016lx) (%4lu%s)\n",
591 i, efi_md_typeattr_format(buf, sizeof(buf), md),
592 md->phys_addr,
593 md->phys_addr + efi_md_size(md), size, unit);
594 }
595 }
596#endif
597
598 efi_map_pal_code();
599 efi_enter_virtual_mode();
600}
601
602void
603efi_enter_virtual_mode (void)
604{
605 void *efi_map_start, *efi_map_end, *p;
606 efi_memory_desc_t *md;
607 efi_status_t status;
608 u64 efi_desc_size;
609
610 efi_map_start = __va(ia64_boot_param->efi_memmap);
611 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
612 efi_desc_size = ia64_boot_param->efi_memdesc_size;
613
614 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
615 md = p;
616 if (md->attribute & EFI_MEMORY_RUNTIME) {
617 /*
618 * Some descriptors have multiple bits set, so the
619 * order of the tests is relevant.
620 */
621 if (md->attribute & EFI_MEMORY_WB) {
622 md->virt_addr = (u64) __va(md->phys_addr);
623 } else if (md->attribute & EFI_MEMORY_UC) {
624 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
625 } else if (md->attribute & EFI_MEMORY_WC) {
626#if 0
627 md->virt_addr = ia64_remap(md->phys_addr,
628 (_PAGE_A |
629 _PAGE_P |
630 _PAGE_D |
631 _PAGE_MA_WC |
632 _PAGE_PL_0 |
633 _PAGE_AR_RW));
634#else
635 printk(KERN_INFO "EFI_MEMORY_WC mapping\n");
636 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
637#endif
638 } else if (md->attribute & EFI_MEMORY_WT) {
639#if 0
640 md->virt_addr = ia64_remap(md->phys_addr,
641 (_PAGE_A |
642 _PAGE_P |
643 _PAGE_D |
644 _PAGE_MA_WT |
645 _PAGE_PL_0 |
646 _PAGE_AR_RW));
647#else
648 printk(KERN_INFO "EFI_MEMORY_WT mapping\n");
649 md->virt_addr = (u64) ioremap(md->phys_addr, 0);
650#endif
651 }
652 }
653 }
654
655 status = efi_call_phys(__va(runtime->set_virtual_address_map),
656 ia64_boot_param->efi_memmap_size,
657 efi_desc_size,
658 ia64_boot_param->efi_memdesc_version,
659 ia64_boot_param->efi_memmap);
660 if (status != EFI_SUCCESS) {
661 printk(KERN_WARNING "warning: unable to switch EFI into "
662 "virtual mode (status=%lu)\n", status);
663 return;
664 }
665
666 set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
667
668 /*
669 * Now that EFI is in virtual mode, we call the EFI functions more
670 * efficiently:
671 */
672 efi.get_time = virt_get_time;
673 efi.set_time = virt_set_time;
674 efi.get_wakeup_time = virt_get_wakeup_time;
675 efi.set_wakeup_time = virt_set_wakeup_time;
676 efi.get_variable = virt_get_variable;
677 efi.get_next_variable = virt_get_next_variable;
678 efi.set_variable = virt_set_variable;
679 efi.get_next_high_mono_count = virt_get_next_high_mono_count;
680 efi.reset_system = virt_reset_system;
681}
682
683/*
684 * Walk the EFI memory map looking for the I/O port range. There can only be
685 * one entry of this type, other I/O port ranges should be described via ACPI.
686 */
687u64
688efi_get_iobase (void)
689{
690 void *efi_map_start, *efi_map_end, *p;
691 efi_memory_desc_t *md;
692 u64 efi_desc_size;
693
694 efi_map_start = __va(ia64_boot_param->efi_memmap);
695 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
696 efi_desc_size = ia64_boot_param->efi_memdesc_size;
697
698 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
699 md = p;
700 if (md->type == EFI_MEMORY_MAPPED_IO_PORT_SPACE) {
701 if (md->attribute & EFI_MEMORY_UC)
702 return md->phys_addr;
703 }
704 }
705 return 0;
706}
707
708static struct kern_memdesc *
709kern_memory_descriptor (unsigned long phys_addr)
710{
711 struct kern_memdesc *md;
712
713 for (md = kern_memmap; md->start != ~0UL; md++) {
714 if (phys_addr - md->start < (md->num_pages << EFI_PAGE_SHIFT))
715 return md;
716 }
717 return NULL;
718}
719
720static efi_memory_desc_t *
721efi_memory_descriptor (unsigned long phys_addr)
722{
723 void *efi_map_start, *efi_map_end, *p;
724 efi_memory_desc_t *md;
725 u64 efi_desc_size;
726
727 efi_map_start = __va(ia64_boot_param->efi_memmap);
728 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
729 efi_desc_size = ia64_boot_param->efi_memdesc_size;
730
731 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
732 md = p;
733
734 if (phys_addr - md->phys_addr < efi_md_size(md))
735 return md;
736 }
737 return NULL;
738}
739
740static int
741efi_memmap_intersects (unsigned long phys_addr, unsigned long size)
742{
743 void *efi_map_start, *efi_map_end, *p;
744 efi_memory_desc_t *md;
745 u64 efi_desc_size;
746 unsigned long end;
747
748 efi_map_start = __va(ia64_boot_param->efi_memmap);
749 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
750 efi_desc_size = ia64_boot_param->efi_memdesc_size;
751
752 end = phys_addr + size;
753
754 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
755 md = p;
756 if (md->phys_addr < end && efi_md_end(md) > phys_addr)
757 return 1;
758 }
759 return 0;
760}
761
762u32
763efi_mem_type (unsigned long phys_addr)
764{
765 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
766
767 if (md)
768 return md->type;
769 return 0;
770}
771
772u64
773efi_mem_attributes (unsigned long phys_addr)
774{
775 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
776
777 if (md)
778 return md->attribute;
779 return 0;
780}
781EXPORT_SYMBOL(efi_mem_attributes);
782
783u64
784efi_mem_attribute (unsigned long phys_addr, unsigned long size)
785{
786 unsigned long end = phys_addr + size;
787 efi_memory_desc_t *md = efi_memory_descriptor(phys_addr);
788 u64 attr;
789
790 if (!md)
791 return 0;
792
793 /*
794 * EFI_MEMORY_RUNTIME is not a memory attribute; it just tells
795 * the kernel that firmware needs this region mapped.
796 */
797 attr = md->attribute & ~EFI_MEMORY_RUNTIME;
798 do {
799 unsigned long md_end = efi_md_end(md);
800
801 if (end <= md_end)
802 return attr;
803
804 md = efi_memory_descriptor(md_end);
805 if (!md || (md->attribute & ~EFI_MEMORY_RUNTIME) != attr)
806 return 0;
807 } while (md);
808 return 0; /* never reached */
809}
810
811u64
812kern_mem_attribute (unsigned long phys_addr, unsigned long size)
813{
814 unsigned long end = phys_addr + size;
815 struct kern_memdesc *md;
816 u64 attr;
817
818 /*
819 * This is a hack for ioremap calls before we set up kern_memmap.
820 * Maybe we should do efi_memmap_init() earlier instead.
821 */
822 if (!kern_memmap) {
823 attr = efi_mem_attribute(phys_addr, size);
824 if (attr & EFI_MEMORY_WB)
825 return EFI_MEMORY_WB;
826 return 0;
827 }
828
829 md = kern_memory_descriptor(phys_addr);
830 if (!md)
831 return 0;
832
833 attr = md->attribute;
834 do {
835 unsigned long md_end = kmd_end(md);
836
837 if (end <= md_end)
838 return attr;
839
840 md = kern_memory_descriptor(md_end);
841 if (!md || md->attribute != attr)
842 return 0;
843 } while (md);
844 return 0; /* never reached */
845}
846EXPORT_SYMBOL(kern_mem_attribute);
847
848int
849valid_phys_addr_range (phys_addr_t phys_addr, unsigned long size)
850{
851 u64 attr;
852
853 /*
854 * /dev/mem reads and writes use copy_to_user(), which implicitly
855 * uses a granule-sized kernel identity mapping. It's really
856 * only safe to do this for regions in kern_memmap. For more
857 * details, see Documentation/ia64/aliasing.txt.
858 */
859 attr = kern_mem_attribute(phys_addr, size);
860 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
861 return 1;
862 return 0;
863}
864
865int
866valid_mmap_phys_addr_range (unsigned long pfn, unsigned long size)
867{
868 unsigned long phys_addr = pfn << PAGE_SHIFT;
869 u64 attr;
870
871 attr = efi_mem_attribute(phys_addr, size);
872
873 /*
874 * /dev/mem mmap uses normal user pages, so we don't need the entire
875 * granule, but the entire region we're mapping must support the same
876 * attribute.
877 */
878 if (attr & EFI_MEMORY_WB || attr & EFI_MEMORY_UC)
879 return 1;
880
881 /*
882 * Intel firmware doesn't tell us about all the MMIO regions, so
883 * in general we have to allow mmap requests. But if EFI *does*
884 * tell us about anything inside this region, we should deny it.
885 * The user can always map a smaller region to avoid the overlap.
886 */
887 if (efi_memmap_intersects(phys_addr, size))
888 return 0;
889
890 return 1;
891}
892
893pgprot_t
894phys_mem_access_prot(struct file *file, unsigned long pfn, unsigned long size,
895 pgprot_t vma_prot)
896{
897 unsigned long phys_addr = pfn << PAGE_SHIFT;
898 u64 attr;
899
900 /*
901 * For /dev/mem mmap, we use user mappings, but if the region is
902 * in kern_memmap (and hence may be covered by a kernel mapping),
903 * we must use the same attribute as the kernel mapping.
904 */
905 attr = kern_mem_attribute(phys_addr, size);
906 if (attr & EFI_MEMORY_WB)
907 return pgprot_cacheable(vma_prot);
908 else if (attr & EFI_MEMORY_UC)
909 return pgprot_noncached(vma_prot);
910
911 /*
912 * Some chipsets don't support UC access to memory. If
913 * WB is supported, we prefer that.
914 */
915 if (efi_mem_attribute(phys_addr, size) & EFI_MEMORY_WB)
916 return pgprot_cacheable(vma_prot);
917
918 return pgprot_noncached(vma_prot);
919}
920
921int __init
922efi_uart_console_only(void)
923{
924 efi_status_t status;
925 char *s, name[] = "ConOut";
926 efi_guid_t guid = EFI_GLOBAL_VARIABLE_GUID;
927 efi_char16_t *utf16, name_utf16[32];
928 unsigned char data[1024];
929 unsigned long size = sizeof(data);
930 struct efi_generic_dev_path *hdr, *end_addr;
931 int uart = 0;
932
933 /* Convert to UTF-16 */
934 utf16 = name_utf16;
935 s = name;
936 while (*s)
937 *utf16++ = *s++ & 0x7f;
938 *utf16 = 0;
939
940 status = efi.get_variable(name_utf16, &guid, NULL, &size, data);
941 if (status != EFI_SUCCESS) {
942 printk(KERN_ERR "No EFI %s variable?\n", name);
943 return 0;
944 }
945
946 hdr = (struct efi_generic_dev_path *) data;
947 end_addr = (struct efi_generic_dev_path *) ((u8 *) data + size);
948 while (hdr < end_addr) {
949 if (hdr->type == EFI_DEV_MSG &&
950 hdr->sub_type == EFI_DEV_MSG_UART)
951 uart = 1;
952 else if (hdr->type == EFI_DEV_END_PATH ||
953 hdr->type == EFI_DEV_END_PATH2) {
954 if (!uart)
955 return 0;
956 if (hdr->sub_type == EFI_DEV_END_ENTIRE)
957 return 1;
958 uart = 0;
959 }
960 hdr = (struct efi_generic_dev_path *)((u8 *) hdr + hdr->length);
961 }
962 printk(KERN_ERR "Malformed %s value\n", name);
963 return 0;
964}
965
966/*
967 * Look for the first granule aligned memory descriptor memory
968 * that is big enough to hold EFI memory map. Make sure this
969 * descriptor is atleast granule sized so it does not get trimmed
970 */
971struct kern_memdesc *
972find_memmap_space (void)
973{
974 u64 contig_low=0, contig_high=0;
975 u64 as = 0, ae;
976 void *efi_map_start, *efi_map_end, *p, *q;
977 efi_memory_desc_t *md, *pmd = NULL, *check_md;
978 u64 space_needed, efi_desc_size;
979 unsigned long total_mem = 0;
980
981 efi_map_start = __va(ia64_boot_param->efi_memmap);
982 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
983 efi_desc_size = ia64_boot_param->efi_memdesc_size;
984
985 /*
986 * Worst case: we need 3 kernel descriptors for each efi descriptor
987 * (if every entry has a WB part in the middle, and UC head and tail),
988 * plus one for the end marker.
989 */
990 space_needed = sizeof(kern_memdesc_t) *
991 (3 * (ia64_boot_param->efi_memmap_size/efi_desc_size) + 1);
992
993 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
994 md = p;
995 if (!efi_wb(md)) {
996 continue;
997 }
998 if (pmd == NULL || !efi_wb(pmd) ||
999 efi_md_end(pmd) != md->phys_addr) {
1000 contig_low = GRANULEROUNDUP(md->phys_addr);
1001 contig_high = efi_md_end(md);
1002 for (q = p + efi_desc_size; q < efi_map_end;
1003 q += efi_desc_size) {
1004 check_md = q;
1005 if (!efi_wb(check_md))
1006 break;
1007 if (contig_high != check_md->phys_addr)
1008 break;
1009 contig_high = efi_md_end(check_md);
1010 }
1011 contig_high = GRANULEROUNDDOWN(contig_high);
1012 }
1013 if (!is_memory_available(md) || md->type == EFI_LOADER_DATA)
1014 continue;
1015
1016 /* Round ends inward to granule boundaries */
1017 as = max(contig_low, md->phys_addr);
1018 ae = min(contig_high, efi_md_end(md));
1019
1020 /* keep within max_addr= and min_addr= command line arg */
1021 as = max(as, min_addr);
1022 ae = min(ae, max_addr);
1023 if (ae <= as)
1024 continue;
1025
1026 /* avoid going over mem= command line arg */
1027 if (total_mem + (ae - as) > mem_limit)
1028 ae -= total_mem + (ae - as) - mem_limit;
1029
1030 if (ae <= as)
1031 continue;
1032
1033 if (ae - as > space_needed)
1034 break;
1035 }
1036 if (p >= efi_map_end)
1037 panic("Can't allocate space for kernel memory descriptors");
1038
1039 return __va(as);
1040}
1041
1042/*
1043 * Walk the EFI memory map and gather all memory available for kernel
1044 * to use. We can allocate partial granules only if the unavailable
1045 * parts exist, and are WB.
1046 */
1047unsigned long
1048efi_memmap_init(u64 *s, u64 *e)
1049{
1050 struct kern_memdesc *k, *prev = NULL;
1051 u64 contig_low=0, contig_high=0;
1052 u64 as, ae, lim;
1053 void *efi_map_start, *efi_map_end, *p, *q;
1054 efi_memory_desc_t *md, *pmd = NULL, *check_md;
1055 u64 efi_desc_size;
1056 unsigned long total_mem = 0;
1057
1058 k = kern_memmap = find_memmap_space();
1059
1060 efi_map_start = __va(ia64_boot_param->efi_memmap);
1061 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1062 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1063
1064 for (p = efi_map_start; p < efi_map_end; pmd = md, p += efi_desc_size) {
1065 md = p;
1066 if (!efi_wb(md)) {
1067 if (efi_uc(md) &&
1068 (md->type == EFI_CONVENTIONAL_MEMORY ||
1069 md->type == EFI_BOOT_SERVICES_DATA)) {
1070 k->attribute = EFI_MEMORY_UC;
1071 k->start = md->phys_addr;
1072 k->num_pages = md->num_pages;
1073 k++;
1074 }
1075 continue;
1076 }
1077 if (pmd == NULL || !efi_wb(pmd) ||
1078 efi_md_end(pmd) != md->phys_addr) {
1079 contig_low = GRANULEROUNDUP(md->phys_addr);
1080 contig_high = efi_md_end(md);
1081 for (q = p + efi_desc_size; q < efi_map_end;
1082 q += efi_desc_size) {
1083 check_md = q;
1084 if (!efi_wb(check_md))
1085 break;
1086 if (contig_high != check_md->phys_addr)
1087 break;
1088 contig_high = efi_md_end(check_md);
1089 }
1090 contig_high = GRANULEROUNDDOWN(contig_high);
1091 }
1092 if (!is_memory_available(md))
1093 continue;
1094
1095 /*
1096 * Round ends inward to granule boundaries
1097 * Give trimmings to uncached allocator
1098 */
1099 if (md->phys_addr < contig_low) {
1100 lim = min(efi_md_end(md), contig_low);
1101 if (efi_uc(md)) {
1102 if (k > kern_memmap &&
1103 (k-1)->attribute == EFI_MEMORY_UC &&
1104 kmd_end(k-1) == md->phys_addr) {
1105 (k-1)->num_pages +=
1106 (lim - md->phys_addr)
1107 >> EFI_PAGE_SHIFT;
1108 } else {
1109 k->attribute = EFI_MEMORY_UC;
1110 k->start = md->phys_addr;
1111 k->num_pages = (lim - md->phys_addr)
1112 >> EFI_PAGE_SHIFT;
1113 k++;
1114 }
1115 }
1116 as = contig_low;
1117 } else
1118 as = md->phys_addr;
1119
1120 if (efi_md_end(md) > contig_high) {
1121 lim = max(md->phys_addr, contig_high);
1122 if (efi_uc(md)) {
1123 if (lim == md->phys_addr && k > kern_memmap &&
1124 (k-1)->attribute == EFI_MEMORY_UC &&
1125 kmd_end(k-1) == md->phys_addr) {
1126 (k-1)->num_pages += md->num_pages;
1127 } else {
1128 k->attribute = EFI_MEMORY_UC;
1129 k->start = lim;
1130 k->num_pages = (efi_md_end(md) - lim)
1131 >> EFI_PAGE_SHIFT;
1132 k++;
1133 }
1134 }
1135 ae = contig_high;
1136 } else
1137 ae = efi_md_end(md);
1138
1139 /* keep within max_addr= and min_addr= command line arg */
1140 as = max(as, min_addr);
1141 ae = min(ae, max_addr);
1142 if (ae <= as)
1143 continue;
1144
1145 /* avoid going over mem= command line arg */
1146 if (total_mem + (ae - as) > mem_limit)
1147 ae -= total_mem + (ae - as) - mem_limit;
1148
1149 if (ae <= as)
1150 continue;
1151 if (prev && kmd_end(prev) == md->phys_addr) {
1152 prev->num_pages += (ae - as) >> EFI_PAGE_SHIFT;
1153 total_mem += ae - as;
1154 continue;
1155 }
1156 k->attribute = EFI_MEMORY_WB;
1157 k->start = as;
1158 k->num_pages = (ae - as) >> EFI_PAGE_SHIFT;
1159 total_mem += ae - as;
1160 prev = k++;
1161 }
1162 k->start = ~0L; /* end-marker */
1163
1164 /* reserve the memory we are using for kern_memmap */
1165 *s = (u64)kern_memmap;
1166 *e = (u64)++k;
1167
1168 return total_mem;
1169}
1170
1171void
1172efi_initialize_iomem_resources(struct resource *code_resource,
1173 struct resource *data_resource,
1174 struct resource *bss_resource)
1175{
1176 struct resource *res;
1177 void *efi_map_start, *efi_map_end, *p;
1178 efi_memory_desc_t *md;
1179 u64 efi_desc_size;
1180 char *name;
1181 unsigned long flags, desc;
1182
1183 efi_map_start = __va(ia64_boot_param->efi_memmap);
1184 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1185 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1186
1187 res = NULL;
1188
1189 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1190 md = p;
1191
1192 if (md->num_pages == 0) /* should not happen */
1193 continue;
1194
1195 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1196 desc = IORES_DESC_NONE;
1197
1198 switch (md->type) {
1199
1200 case EFI_MEMORY_MAPPED_IO:
1201 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
1202 continue;
1203
1204 case EFI_LOADER_CODE:
1205 case EFI_LOADER_DATA:
1206 case EFI_BOOT_SERVICES_DATA:
1207 case EFI_BOOT_SERVICES_CODE:
1208 case EFI_CONVENTIONAL_MEMORY:
1209 if (md->attribute & EFI_MEMORY_WP) {
1210 name = "System ROM";
1211 flags |= IORESOURCE_READONLY;
1212 } else if (md->attribute == EFI_MEMORY_UC) {
1213 name = "Uncached RAM";
1214 } else {
1215 name = "System RAM";
1216 flags |= IORESOURCE_SYSRAM;
1217 }
1218 break;
1219
1220 case EFI_ACPI_MEMORY_NVS:
1221 name = "ACPI Non-volatile Storage";
1222 desc = IORES_DESC_ACPI_NV_STORAGE;
1223 break;
1224
1225 case EFI_UNUSABLE_MEMORY:
1226 name = "reserved";
1227 flags |= IORESOURCE_DISABLED;
1228 break;
1229
1230 case EFI_PERSISTENT_MEMORY:
1231 name = "Persistent Memory";
1232 desc = IORES_DESC_PERSISTENT_MEMORY;
1233 break;
1234
1235 case EFI_RESERVED_TYPE:
1236 case EFI_RUNTIME_SERVICES_CODE:
1237 case EFI_RUNTIME_SERVICES_DATA:
1238 case EFI_ACPI_RECLAIM_MEMORY:
1239 default:
1240 name = "reserved";
1241 break;
1242 }
1243
1244 if ((res = kzalloc(sizeof(struct resource),
1245 GFP_KERNEL)) == NULL) {
1246 printk(KERN_ERR
1247 "failed to allocate resource for iomem\n");
1248 return;
1249 }
1250
1251 res->name = name;
1252 res->start = md->phys_addr;
1253 res->end = md->phys_addr + efi_md_size(md) - 1;
1254 res->flags = flags;
1255 res->desc = desc;
1256
1257 if (insert_resource(&iomem_resource, res) < 0)
1258 kfree(res);
1259 else {
1260 /*
1261 * We don't know which region contains
1262 * kernel data so we try it repeatedly and
1263 * let the resource manager test it.
1264 */
1265 insert_resource(res, code_resource);
1266 insert_resource(res, data_resource);
1267 insert_resource(res, bss_resource);
1268#ifdef CONFIG_KEXEC
1269 insert_resource(res, &efi_memmap_res);
1270 insert_resource(res, &boot_param_res);
1271 if (crashk_res.end > crashk_res.start)
1272 insert_resource(res, &crashk_res);
1273#endif
1274 }
1275 }
1276}
1277
1278#ifdef CONFIG_KEXEC
1279/* find a block of memory aligned to 64M exclude reserved regions
1280 rsvd_regions are sorted
1281 */
1282unsigned long __init
1283kdump_find_rsvd_region (unsigned long size, struct rsvd_region *r, int n)
1284{
1285 int i;
1286 u64 start, end;
1287 u64 alignment = 1UL << _PAGE_SIZE_64M;
1288 void *efi_map_start, *efi_map_end, *p;
1289 efi_memory_desc_t *md;
1290 u64 efi_desc_size;
1291
1292 efi_map_start = __va(ia64_boot_param->efi_memmap);
1293 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1294 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1295
1296 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1297 md = p;
1298 if (!efi_wb(md))
1299 continue;
1300 start = ALIGN(md->phys_addr, alignment);
1301 end = efi_md_end(md);
1302 for (i = 0; i < n; i++) {
1303 if (__pa(r[i].start) >= start && __pa(r[i].end) < end) {
1304 if (__pa(r[i].start) > start + size)
1305 return start;
1306 start = ALIGN(__pa(r[i].end), alignment);
1307 if (i < n-1 &&
1308 __pa(r[i+1].start) < start + size)
1309 continue;
1310 else
1311 break;
1312 }
1313 }
1314 if (end > start + size)
1315 return start;
1316 }
1317
1318 printk(KERN_WARNING
1319 "Cannot reserve 0x%lx byte of memory for crashdump\n", size);
1320 return ~0UL;
1321}
1322#endif
1323
1324#ifdef CONFIG_CRASH_DUMP
1325/* locate the size find a the descriptor at a certain address */
1326unsigned long __init
1327vmcore_find_descriptor_size (unsigned long address)
1328{
1329 void *efi_map_start, *efi_map_end, *p;
1330 efi_memory_desc_t *md;
1331 u64 efi_desc_size;
1332 unsigned long ret = 0;
1333
1334 efi_map_start = __va(ia64_boot_param->efi_memmap);
1335 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
1336 efi_desc_size = ia64_boot_param->efi_memdesc_size;
1337
1338 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
1339 md = p;
1340 if (efi_wb(md) && md->type == EFI_LOADER_DATA
1341 && md->phys_addr == address) {
1342 ret = efi_md_size(md);
1343 break;
1344 }
1345 }
1346
1347 if (ret == 0)
1348 printk(KERN_WARNING "Cannot locate EFI vmcore descriptor\n");
1349
1350 return ret;
1351}
1352#endif